Reproduce “Generative AI meets 3D: A Survey on Text-to-3D in AIGC Era” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Chenghao Li, Chaoning Zhang, Joseph Cho, Atish Waghwase, Lik-Hang Lee, Francois Rameau, Yang Yang, Sung-Ho Bae and 1 more
A comprehensive survey maps the rapidly evolving text-to-3D generation landscape, systematically categorizing 3D data representations, foundational technologies, and methods. It outlines key challenges, enhancement directions in fidelity, efficiency, consistency, controllability, and diversity, and diverse applications across industries. The paper also provides a future agenda for responsible and holistic development of text-to-3D technology.
Key insights
- Neural Fields (e.g., NeRF, SDF) are identified as the most promising 3D data representations for generative tasks due to their high representational capability and memory efficiency.
- Modern text-to-3D generation is driven by the convergence of foundational technologies: Neural Radiance Fields (NeRF), diffusion models, unified text-vision representations (CLIP), and Score Distillation Sampling (SDS).
- Research efforts address five critical enhancement directions: improving fidelity, increasing efficiency, ensuring consistency, enhancing controllability, and promoting diversity in generated 3D content.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.